Nonlinear System Identification with Multiple Data Sets for Structures with Bolted Joints

Often, system identification is performed on a single data set at a time. When multiple data sets exist, an approach to considering them is to analyze the resulting identified parameters statistically (such as the average frequency, or 95% confidence interval of extracted parameters, etc.). An alternative could be a method to identify the parameters of a system based on data from multiple measurements; then this would potentially lead to an identified system model that is valid over a much larger operating range. In this chapter, new methods are investigated for the estimation of nonlinear characteristics when large amounts of data are available. The methods include direct nonlinear optimization–based identification techniques like the more commonly used sparse identification package, SINDy, and a more customizable sequential learning algorithm. Besides, parameter initialization (such that any optimized models are able to avoid bad local minima) is studied to accomplish a successful identification. Multiple data sets from an experimental setup of nonlinear structures with the expected source of nonlinearity, i.e., the bolted joints are used in this study to evaluate the performance of the identification methods. For assessing the quality of the resulting models, the responses from simulations are compared to the measured responses of the structures such as amplitude-dependent frequencies and damping ratios.